Verdi is a Coq framework to implement and formally verify distributed systems. Verdi supports several different fault models ranging from idealistic to realistic. Verdi's verified system transformers (VSTs) encapsulate common fault tolerance techniques. Developers can verify an application in an idealized fault model, and then apply a VST to obtain an application that is guaranteed to have analogous properties in a more adversarial environment.
- Author(s):
- Justin Adsuara
- Steve Anton
- Ryan Doenges
- Karl Palmskog
- Pavel Panchekha
- Zachary Tatlock
- James R. Wilcox
- Doug Woos
- License: BSD 2-Clause "Simplified" license
- Compatible Coq versions: 8.14 or later
- Additional dependencies:
- Coq namespace:
Verdi
- Related publication(s):
We recommend installing Verdi via opam, which will automatically build and install its dependencies:
opam repo add coq-extra-dev https://coq.inria.fr/opam/extra-dev
opam install coq-verdi
To build Verdi manually, first install all requirements. Then,
run make
in the Verdi root directory. This will compile
the framework's core specifications and proofs, as well as some
simple example systems and their correctness proofs.
To run Verdi systems on real hardware, event handler code must be extracted to OCaml and linked with one of the shims in the Verdi runtime library that handles low-level network communication.
To set up your own Verdi-based distributed systems verification project, we recommend basing it on Verdi LockServ.
Verdi LockServ contains a minimalistic implementation of a message-passing lock server and a proof that it maintains mutual exclusion between client nodes. At build time, extracted OCaml code is linked to a runtime library shim to produce an executable program that can be run in a cluster. There is also a simple script to interface with cluster nodes.
In addition to the example verified systems listed below, see the scientific papers and blog posts listed at the Verdi website. See also Verdi Raft, a verified implementation of the Raft distributed consensus protocol.
- Core Verdi files:
Verdi.v
: exporting of core Verdi theories, imported by systemsNet.v
: core (unlabeled) network semanticsLabeledNet.v
: labeled network semantics, for use in liveness reasoningHandlerMonad.v
: a monad for writing network/input handlersStatePacketPacket.v
: a technique for writing easily decomposable invariants
- Example systems:
Counter.v
: counting server with backupLockServ.v
: lock server with proof of safetyLiveLockServ.v
: lock server with proof of livenessVarD.v
:vard
, a key-value store
- Verified system transformers:
SeqNum.v
andSeqNumCorrect.v
, a system transformer implementing sequence numberingLockServSeqNum.v
, the sequence numbering transformer applied to the lock serverPrimaryBackup.v
, a system transformer implementing asynchronous primary-backup replicationVarDPrimaryBackup.v
, the primary-backup transformer applied to the key-value store